Automated hierarchical tuning of configuration parameters for a multi-layer service

A non-transitory machine readable medium storing instructions executable by a processing resource of a computer system, the non-transitory machine readable medium comprising instructions configurable to cause the processing resource to:

• process a service definition and optimization criteria for tuning a configuration of a service for a particular production environment, wherein the service definition includes information regarding a plurality of layers of the service and corresponding configuration groups, each configuration group including configuration parameters for a corresponding layer of the service and for which values are configurable;
• limit the configuration parameters to those configuration parameters represented within the corresponding configuration groups that have a particular level of an impact on an optimization goal for the service;
• create an acyclic dependency graph including a plurality of nodes, wherein a node represents a layer in the plurality of layers, the acyclic dependency graph representing the corresponding configuration groups passed between the plurality of layers and representing a hierarchical relationship between the plurality of layers of the service;
• and
• globally optimize values of configuration parameters of the corresponding configuration groups by:
• creating an instance of the service based on the service definition;
• and
• performing a local optimization process based on the optimization criteria at each layer of the plurality of layers of the instance of the service by locally optimizing values of configuration parameters associated with a particular layer in the plurality of layers for a first node before passing the optimized values of configuration parameters for the first node on to a parent node associated with a parent layer in the plurality of layers as defined by the acyclic dependency graph and propagating respective optimized values for respective nodes that are associated with respective layers through the nodes of the dependency graph.

A system comprising:

• a processing resource;
• a non-transitory computer-readable medium, coupled to the processing resource, having stored therein instructions that are configurable to cause the processing resource to:
• process a service definition and optimization criteria for tuning a configuration of a service for a particular production environment, wherein the service definition includes information regarding a plurality of layers of the service and corresponding configuration groups, each configuration group including configuration parameters for a corresponding layer of the service and for which values are configurable;
• limit the configuration parameters to those configuration parameters represented within the corresponding configuration groups that have a particular level of an impact on an optimization goal for the service;
• create an acyclic dependency graph including a plurality of nodes, wherein a node represents a layer in the plurality of layers, the acyclic dependency graph representing configuration groups passed between the plurality of layers and representing a hierarchical relationship between the plurality of layers of the service;
• and
• globally optimize values of configuration parameters of the configuration groups by:
• creating an instance of the service based on the service definition;
• and
• until a stopping condition is achieved, iteratively performing a local optimization process based on the optimization criteria at each layer of the plurality of layers of the instance of the service by locally optimizing value of configuration parameters associated with a particular layer in the plurality of layers for a first node before passing the optimized values of configuration parameters for the first node on to a parent node associated with a parent layer in the plurality of layers as defined by the acyclic dependency graph and propagating respective optimized values for respective nodes that are associated with respective layers through the nodes of the dependency graph.